fix: snapshot AI streaming preview and task PR flow

.changeset/ai-preview-history.md

@@ -0,0 +1,5 @@
+---
+"@platejs/ai": patch
+---
+
+Fix insert-mode AI preview history so streamed chunks stay out of undo history and selection survives accept, undo, and redo

.changeset/slate-redo-selection.md

@@ -0,0 +1,5 @@
+---
+"@platejs/slate": patch
+---
+
+Fix redo to restore selection after undoing history batches that clear the active selection

.claude/.skiller.json

@@ -26,6 +26,22 @@
         "sourceRelPath": ".claude/commands/translate.md",
         "destRelPath": "translate"
       },
+      {
+        "sourceType": "plugin",
+        "pluginId": "compound-engineering@every-marketplace",
+        "pluginVersion": "2.40.0",
+        "sourceKind": "agent",
+        "sourceRelPath": "agents/document-review/adversarial-document-reviewer.md",
+        "destRelPath": "adversarial-document-reviewer"
+      },
+      {
+        "sourceType": "plugin",
+        "pluginId": "compound-engineering@every-marketplace",
+        "pluginVersion": "2.40.0",
+        "sourceKind": "agent",
+        "sourceRelPath": "agents/review/adversarial-reviewer.md",
+        "destRelPath": "adversarial-reviewer"
+      },
       {
         "sourceType": "plugin",
         "pluginId": "agent-browser@agent-browser",
@@ -199,24 +215,24 @@
         "pluginId": "compound-engineering@every-marketplace",
         "pluginVersion": "2.40.0",
         "sourceKind": "agent",
-        "sourceRelPath": "agents/review/code-simplicity-reviewer.md",
-        "destRelPath": "code-simplicity-reviewer"
+        "sourceRelPath": "agents/review/cli-agent-readiness-reviewer.md",
+        "destRelPath": "cli-agent-readiness-reviewer"
       },
       {
         "sourceType": "plugin",
         "pluginId": "compound-engineering@every-marketplace",
         "pluginVersion": "2.40.0",
         "sourceKind": "agent",
-        "sourceRelPath": "agents/document-review/coherence-reviewer.md",
-        "destRelPath": "coherence-reviewer"
+        "sourceRelPath": "agents/review/code-simplicity-reviewer.md",
+        "destRelPath": "code-simplicity-reviewer"
       },
       {
         "sourceType": "plugin",
         "pluginId": "compound-engineering@every-marketplace",
         "pluginVersion": "2.40.0",
-        "sourceKind": "skill",
-        "sourceRelPath": "skills/compound-docs",
-        "destRelPath": "compound-docs"
+        "sourceKind": "agent",
+        "sourceRelPath": "agents/document-review/coherence-reviewer.md",
+        "destRelPath": "coherence-reviewer"
       },
       {
         "sourceType": "plugin",
@@ -282,14 +298,6 @@
         "sourceRelPath": "skills/debug",
         "destRelPath": "debug"
       },
-      {
-        "sourceType": "plugin",
-        "pluginId": "compound-engineering@every-marketplace",
-        "pluginVersion": "2.40.0",
-        "sourceKind": "skill",
-        "sourceRelPath": "skills/deepen-plan",
-        "destRelPath": "deepen-plan"
-      },
       {
         "sourceType": "plugin",
         "pluginId": "compound-engineering@every-marketplace",
@@ -626,6 +634,14 @@
         "sourceRelPath": "agents/document-review/product-lens-reviewer.md",
         "destRelPath": "product-lens-reviewer"
       },
+      {
+        "sourceType": "plugin",
+        "pluginId": "compound-engineering@every-marketplace",
+        "pluginVersion": "2.40.0",
+        "sourceKind": "agent",
+        "sourceRelPath": "agents/review/project-standards-reviewer.md",
+        "destRelPath": "project-standards-reviewer"
+      },
       {
         "sourceType": "plugin",
         "pluginId": "compound-engineering@every-marketplace",

.claude/docs/plans/2026-03-26-ai-preview-localized-rollback.md

@@ -0,0 +1,53 @@
+# AI Preview Localized Rollback
+
+## Goal
+
+Replace full-document AI preview snapshotting with localized insert-mode rollback so preview accept/cancel only touch the preview block range, while preserving current undo/redo and selection behavior.
+
+## Checklist
+
+- [completed] Read current localized stream path, accept path, and preview tests
+- [completed] Add failing tests for localized preview ownership and rollback
+- [completed] Implement localized preview state and marker-based range handling
+- [completed] Run targeted verification for touched packages and tests
+- [blocked] Run `check`, then create or update the PR
+
+## Verification
+
+- `bun test packages/ai/src/lib/transforms/aiStreamSnapshot.spec.ts packages/ai/src/lib/transforms/undoAI.spec.ts apps/www/src/__tests__/package-integration/ai-chat-streaming/streamHistory.slow.tsx`
+- `pnpm install`
+- `pnpm turbo build --filter=./packages/ai --filter=./apps/www`
+- `pnpm turbo typecheck --filter=./packages/ai`
+- `pnpm build`
+- `pnpm turbo typecheck --filter=./packages/ai --filter=./apps/www` after root build
+- `pnpm lint:fix`
+- `agent-browser open http://localhost:3100/blocks/editor-ai`
+
+## Blocker
+
+- `pnpm check` still fails outside this diff in existing fast AI chat tests:
+  - `packages/ai/src/react/ai-chat/hooks/useAIChatEditor.spec.tsx`
+  - `packages/ai/src/react/ai-chat/hooks/useEditorChat.spec.tsx`
+  - `packages/ai/src/react/ai-chat/utils/getLastAssistantMessage.spec.ts`
+- The failure is a `SyntaxError` from missing exports in `packages/plate/src/react/index.tsx`, so PR creation is blocked by repo rules.
+
+## Agreed Design
+
+- Keep the public surface on `tf.ai.*`
+- Optimize insert-mode preview only
+- Keep preview state private and editor-scoped
+- `beginPreview` accepts `{ originalBlocks }`
+- `selectionBefore` is still captured inside `beginPreview`
+- Preview-owned content is tracked by transient top-level block markers
+- `cancelPreview` restores the exact original block slice plus selection
+- `acceptPreview` localizes cleanup/commit and never rebuilds the whole document
+- Untouched blocks should keep identity because preview accept/cancel stop using full-doc `setValue`
+
+## Open Implementation Notes
+
+- Mark preview-owned top-level blocks with `aiPreview: true`
+- Do not mark the `aiChat` anchor block as preview-owned
+- If preview starts by removing an empty paragraph, capture that exact block in `originalBlocks`
+- If preview inserts after existing content, capture `[]`
+- Derive the current preview range from contiguous marked top-level blocks
+- If preview range is invalid/non-contiguous, fall back safely instead of silently guessing

.claude/docs/plans/2026-03-26-ai-preview-tf-api-refactor.md

@@ -0,0 +1,22 @@
+# AI Preview `tf.ai` Refactor
+
+## Goal
+
+Replace the low-level AI preview snapshot helper surface with `tf.ai.*` lifecycle transforms, keep the full-document snapshot strategy private, migrate insert-mode preview callers, and preserve the current no-history preview behavior.
+
+## Checklist
+
+- [completed] Gather current AI preview/history patterns and relevant learnings
+- [completed] Add or update tests for the new preview lifecycle contract
+- [completed] Implement `tf.ai.*` preview transforms in `BaseAIPlugin`
+- [completed] Migrate AI chat preview callsites off direct snapshot helpers
+- [completed] Run required verification for `packages/ai` and `apps/www`
+
+## Findings
+
+- The current preview storage lives in `packages/ai/src/lib/transforms/aiStreamSnapshot.ts` as a `WeakMap` keyed by editor.
+- The current low-level helper surface is used directly by `acceptAIChat`, `undoAI`, `resetAIChat`, the editor AI kit, and the streaming integration test.
+- The existing solution doc already established the correct behavior boundary: preview chunks stay out of history; accept commits one fresh batch; AI undo restores the pre-stream value while preview is active.
+- `AIPlugin` already owns editor mutation semantics like `insertNodes`, `removeMarks`, `removeNodes`, and `undo`, so preview lifecycle belongs there too.
+- The cleanest public surface here is `tf.ai.*`, while library internals and stricter callsites can still reach the same transforms through `editor.getTransforms(BaseAIPlugin).ai` where generic editor typing is narrower.
+- Package-scoped verification passed for `@platejs/ai`; filtered `apps/www` typecheck still has unrelated workspace export/type failures after the required root `pnpm build`, so that debt remains outside this refactor.

.claude/docs/plans/2026-03-26-ai-stream-snapshot-design-grill.md

@@ -0,0 +1,17 @@
+# AI Stream Snapshot Design Grill
+
+## Goal
+
+Stress-test the next-step design for AI insert-mode preview state, especially whether snapshot state should live in plugin state, what the public surface should be, and how to avoid a performance regression on long documents.
+
+## Checklist
+
+- [in_progress] Inspect current AI/plugin state patterns in the repo
+- [pending] Decide whether preview state should live in plugin state, transform state, or private module state
+- [pending] Decide whether the public surface should be `api`, `tf`, or package-private helpers
+- [pending] Walk the remaining design forks with explicit recommendations
+
+## Findings
+
+- The current implementation stores preview state in a module-level `WeakMap` keyed by editor.
+- The current helper surface is low-level snapshot terminology, not workflow terminology.

.claude/docs/plans/2026-03-26-ignore-tmp-artifacts.md

@@ -0,0 +1,18 @@
+# Ignore Tmp Artifacts
+
+## Goal
+
+Stop root-level `tmp*` artifacts from being committed and remove the accidentally committed browser proof PNGs from the current PR branch.
+
+## Checklist
+
+- [in_progress] Inspect current ignore rules and tracked tmp artifacts
+- [pending] Add ignore rule for root `tmp*` artifacts
+- [pending] Remove tracked tmp PNGs from the branch diff
+- [pending] Run the PR gate
+- [pending] Commit and push the cleanup
+
+## Findings
+
+- `.gitignore` already ignores `/tmp`, but not root files like `tmp-issue-4900-editor-ai.png`.
+- The current branch still tracks `tmp-issue-4900-editor-ai.png` and `tmp-pr-4902-editor-ai.png`.

.claude/docs/plans/2026-03-26-issue-4900-ai-streaming-history.md

@@ -0,0 +1,42 @@
+# Issue 4900: AI streaming history bloat
+
+## Source of truth
+
+- Source type: GitHub issue
+- Source id: #4900
+- Title: Streaming with withAIBatch accumulates operations and may slow undo
+- URL: https://github.com/udecode/plate/issues/4900
+- Task type: bug / performance
+
+## Expected outcome
+
+- Long AI streaming sessions should not build a giant undo batch that makes `tf.ai.undo()` or accept flows slower as chunk count grows.
+- Keep current AI insert/chat behavior intact from a user point of view.
+
+## Constraints and repo rules
+
+- Non-trivial task: use repo-local planning file, not root planning files.
+- Check institutional learnings before implementation.
+- Use a sane test seam before the fix if possible.
+- If `.ts` changes, verify with install -> build -> typecheck sequence for touched package(s), then `pnpm lint:fix`.
+
+## Findings
+
+- No relevant prior solution found in `.claude/docs/solutions/`.
+- `withAIBatch` only merges batches and tags the last undo batch. It does not compact operations.
+- Streaming insert mode is wrapped with `withAIBatch(..., { split: isFirst })` in the AI kit integration.
+- `undoAI` currently relies on the last undo batch being tagged as AI and calls native `editor.undo()`.
+- Accept/hide logic already distinguishes AI preview state from accepted content using AI marks and anchor cleanup.
+
+## Working hypothesis
+
+- The real bug is treating streamed preview updates as normal history.
+- Better fix: keep preview streaming out of history, then finalize or discard preview explicitly.
+- If preview is unsaved, accept likely needs to remove preview without saving, then insert the final accepted content as one normal batch.
+
+## Next steps
+
+1. Confirm accept/reject/reset paths needed for unsaved preview.
+2. Add a failing test that proves streaming no longer bloats undo history.
+3. Implement the smallest package-level fix.
+4. Verify targeted tests, then package build -> typecheck -> lint.

.claude/docs/plans/2026-03-26-pr-and-comment-4900.md

@@ -0,0 +1,24 @@
+# PR And Comment 4900
+
+## Goal
+
+Open or update the PR for the current checkout, then update the existing GitHub issue comment for issue `#4900` with the PR URL.
+
+## Checklist
+
+- [completed] Load the PR workflow and record the task plan
+- [completed] Run `check`
+- [completed] Inspect branch and PR state
+- [completed] Commit and push the full current checkout
+- [completed] Open or update the PR
+- [completed] Update the existing `#4900` issue comment with the PR link
+
+## Findings
+
+- Repo rules require `check` before any PR create or update.
+- Repo rules also require using the entire current checkout as-is, including unrelated changes.
+- The first `check` run failed in `packages/ai/src/react/ai-chat/utils/aiChatActions.spec.ts` because `resetAIChat.ts` imported `clearAIStreamSnapshot` through the AI lib barrel, which pulled in markdown serializer exports the test mock did not provide.
+- Narrowing that import to `../../../lib/transforms/aiStreamSnapshot` fixed the failing spec without changing behavior.
+- `pnpm check` passed after the import fix.
+- Opened PR: `https://github.com/udecode/plate/pull/4902`
+- Updated issue comment: `https://github.com/udecode/plate/issues/4900#issuecomment-4136090707`

.claude/docs/plans/2026-03-26-task-skill-auto-pr-sync.md

@@ -0,0 +1,19 @@
+# Task Skill Auto PR Sync
+
+## Goal
+
+Update the task skill so verified code-changing tracked work opens or updates the PR before any tracker comment, and any GitHub issue comment includes the PR URL.
+
+## Checklist
+
+- [completed] Read the current `task` skill source and adjacent skill/repo rules
+- [completed] Update `.claude/skills/task/task.mdc` with PR-before-comment behavior
+- [completed] Run the skill sync step
+- [completed] Verify the generated task skill reflects the new rules
+
+## Findings
+
+- The current task skill explicitly says not to default to PR creation.
+- The current GitHub sync-back section says not to mention PR mechanics in the issue comment.
+- Repo rules already require `check` before PR creation, so the task skill should defer to that workflow instead of inventing a new one.
+- `pnpm run prepare` runs `pnpm dlx skiller@latest apply`, which regenerated `.codex/skills/task/SKILL.md` with the new wording.

.claude/docs/plans/2026-03-27-fix-ci-pr-4902.md

@@ -0,0 +1,29 @@
+# Fix CI PR 4902
+
+## Goal
+
+Get the current PR back to green by fixing the active CI failure instead of
+guessing at old logs.
+
+## Checklist
+
+- [in_progress] Fetch the current failing CI checks and local learnings
+- [pending] Reproduce the failure locally
+- [pending] Implement the minimal fix
+- [pending] Run targeted verification and the full PR gate
+- [pending] Update the PR branch
+
+## Notes
+
+- User asked to "fix ci" without a specific failure, so the live PR checks are
+  the source of truth.
+- Current failing run is `CI` for commit `1a7eeb74a`.
+- The failure is `pnpm check` -> `pnpm test:slowest`, not lint, build, or
+  typecheck.
+- GitHub runner flagged four AI chat specs over the fast-suite per-test
+  threshold:
+  - `packages/ai/src/react/ai-chat/hooks/useAIChatEditor.spec.tsx`
+  - `packages/ai/src/react/ai-chat/hooks/useEditorChat.spec.tsx`
+  - `packages/ai/src/react/ai-chat/utils/getLastAssistantMessage.spec.ts`
+  - `packages/ai/src/react/ai-chat/utils/submitAIChat.spec.ts`
+- The CI log explicitly says to move those specs to `*.slow.ts[x]`.